convolutional neural networks

Laccase catalyzes oxidation of lignin and aromatic compound with similar structure to this one. Their low substrate specificity results on degradation of similar phenolic compounds. In this context, Molecular Docking was performed with different ligands suggesting potential bio-degradation. Binding active-sites prediction of fungal laccase (access number uniprotkb: A0A166P2X0), from Ganoderma weberianum was performed using machine learning algorithm based on Deep Convolutional Neural Networks (DeepSite-CNNs chemoinformatic tool). Herein, ligands like 2,4 - dichlorophenol, benzidine, sulfisoxazole, trimethoprim and tetracycline were analyzed and two additional reference controls which were 2,2 – azinobis 3 – ethylbenzothiazoline – 6 - sulfonic acid (ABTS) and 2,6 - dimetoxyphenol (2,6 DMP) were used in comparison with the other former mentioned ligands based on high laccase affinity. The five ligands were carried out because their potential biotechnological interest: the antibiotics sulfisoxazole, trimethoprim and tetracycline, and xenobiotics 2,4 - dichlorophenol and benzidine. Molecular docking experiments returned Gibbs free energy of binding (FEB or affinity) for laccase-ligand complexes. The best docking binding-interaction from each laccase-ligand conformation complexes suggest great ability of these ligands to interact with the laccase active-binding site. Herein, FEB values (kcal/mol) were obtained with higher affinity values for reference controls like 2,6 - dimethoxyphenol with -4.8 Kcal/mol and ABTS with -7.1 Kcal/mol. Furthermore, the FEB values were -4.7, -6.5, -6.8, -5.2 and -6.5 Kcal/mol, for 2,4 - dichlorophenol, benzidine, sulfisoxazole, tetracycline and trimethoprim respectively with high prevalence of hydrophobic interaction with functional laccase binding residues. Lastly, this study presents for first time at the bioinformatics field a molecular docking approach for the prediction of potential substrate of laccase from Ganoderma weberianum towards biotechnological application.

Throughout global efforts to defend against the spread of COVID-19 from late 2019 up until now, one of the most crucial factors that has helped combat the pandemic is the development of various screening methods to detect the presence of COVID-19 as conveniently and accurately as possible. One of such methods is the utilization of chest X-Rays (CXRs) to detect anomalies that are concurrent with a patient infected with COVID-19. While yielding results much faster than the traditional RT-PCR test, CXRs tend to be less accurate. Realizing this issue, in our research, we investigated the applications of computer vision in order to better detect COVID-19 from CXRs. Coupled with an extensive image database of CXRs of healthy patients, patients with non-COVID-19 induced pneumonia, and patients positive with COVID-19, convolutional neural networks (CNNs) prove to possess the ability to easily and accurately identify whether or not a patient is infected with COVID-19 in a matter of seconds. Borrowing and adjusting the architectures of three well-tested CNNs: VGG-16, ResNet50, and MobileNetV2, we performed transfer learning and trained three of our own models, then compared and contrasted their differing precisions, accuracies, and efficiencies in correctly labeling patients with and without COVID-19. In the end, all of our models were able to accurately categorize at least 94% of the CXRs, with some performing better than the others; these differences in performance were largely due to the contrasting architectures each of our models borrowed from the three respective CNNs.